Pre-export R&D, Exporting and Productivity Gains – Evidence from ...

CHINA TRADE RESEARCH GROUP

CTRG WORKING PAPER SERIES

Pre-export R&D, Exporting and Productivity Gains – Evidence from Chinese Firms Mi Dai Peking University

Miaojie Yu Peking University

Working Paper 2011/011

© 2011 by Mi Dai and Miaojie Yu. All rights reserved.

Pre-export R&D, Exporting and Productivity Gains —— Evidence from Chinese Firms Mi Dai1

Miaojie Yu2

ABSTRACT In this paper, we estimate the instantaneous and long-run effects of exporting on the productivity of Chinese manufacturing firms during 2001 and 2007. We argue that pre-export R&D plays a crucial role in generating post-entry productivity gains by increasing firms’ absorptive capacity. Adopting propensity score matching in our estimation, we find that: (1) On average, starting to export has an instantaneous effect in raising productivity by 2%, but there are no significant long-run effects.

(2) Firms that have ever invested in R&D before exporting experience large

and lasting productivity gains, while firms without pre-export R&D don’t have productivity gains even instantaneously. (3) The productivity gains of exporting are increasing in the number of years a firm invests in pre-export R&D.

Key words: Export, TFP, Pre-export R&D JEL classification: F1 L1 D24

1

National School of Development, China Center for Economic Research(CCER), Peking University, Beijing 100871, China, [email protected] 2 National School of Development, China Center for Economic Research(CCER), Peking University, Beijing 100871, China, [email protected] 1

1. Introduction One of the most established empirical facts in the recent firm level trade literature is that exporters are more productive. Two hypotheses are consistent with this fact. The first is that more productive firms select into export.Since exporting incurs tremendous fixed cost, such as expenditures on conducting market survey, establishing foreign sales networks and modifying the products to meet the foreign preference, only the relative productive firms are able to recover these costs with their revenues earned overseas(Melitz, 2003). The second hypothesis is often called ―Learning by Exporting‖, which states that exporting can raise firms’ post-entry productivity. These productivity gains can stem from learning from international customers, suppliers and competitors or from confronting fiercer competition in the foreign market. Since the seminal work of Bernard and Jenson (1999), these two hypotheses have been investigated extensively using firm-level datasets from various countries.3 While ―selection into export‖ has been confirmed in almost every previous study, the evidence for ―learning by exporting‖ is quite mixed. Martins et. al (2009) summaries the results of 33 studies that consider the productivity effects of exporting. They report that the number of studies which find positive productivity effects versus the number of studies which find no significant effects are 18 versus 15. Of course, the mixed evidence are partially due to econometric methodology differences. However, even studies with similar estimation methods still reach qualitatively different conclusions.4 This suggests there might be other factors, especially 3

Influential papers include Clerides, et. al (1998) for Columbia, Mexico and Morocco; De Loecker(2007) for

Slovenia; Greenaway and Kneller (2004) , Greenaway and Yu (2004) , Greenaway and Kneller (2008) for UK; Alvarez and López(2005) for Chile, Greenaway et al (2005) for Sweden. see Martins et. al (2009) for a review. 4

Examples include Greenaway et al. (2004) and Arnold and Hussinger(2005), which both adopted matching

2

some other economic variables, that are affecting the productivity gains from exporting, but are ignored by the researchers. If the values of these variables vary by country or by firms sampled, studies based on different countries or even different samples within the same country can come to different conclusions. Previous studies have made some attempts in finding which variables are affecting the productivity gains from exporting. For example, Greenaway and Kneller(2007) finds the competition of the industry in which export entry occurs helps to explain the difference in post-entry productivity gains, and the competition or learning effects are lower in industries where competition is already strong in the domestic market. De Loecker(2007) finds export destination matters: firms who export to richer and more developed countries may have better chance to get access to advanced technologies and managerial experience during their interactions with the foreign customers and suppliers, and will therefore experience more productivity gains from exporting. However, compared with the large amount of literature which estimates the productivity gains from exporting, there are still few papers analyzing the factors that are affecting these productivity gains. In this paper, we argue that pre-export R&D is an important factor affecting the productivity gains from exporting. Intentional and persistent pre-export R&D helps to build absorptive capacity of the firm, i.e. its ability to value, assimilate and exploit external knowledge, and therefore increase the efficiency of learning when a firm is exposed to foreign advanced technologies and managerial experience. Cohen and Levinthal(1989) made the point in their seminal paper that R&D not only directly generates new information and

techniques but find qualitatively different results about learning by exporting. Also see Hahn(2004) and Bernard and Jenson(1999), which both used OLS.

3

innovation(the innovative effect), but also develops the firm’s ability to identify, assimilate and exploit knowledge from the environment(the learning effect). Extensive empirical studies since then have found supportive evidence of the role of R&D in increasing absorptive capacity. For example, Kinoshita(2000) finds that the effect of R&D in increasing firm’s absorptive capacity is more important than the innovative effect for Czech manufacturing firms during 1995 and 1998; Griffith et.al (2004) find R&D to be statistically and economically important in both technological catch-up and innovation using a panel of industries across twelve OECD countries; Hu et al.(2005) finds that in-house R&D significantly complements technology transfer for a panel of Chinese manufacturing firms. All of these studies suggest that the firms who have persistently invested in pre-export R&D activities have more absorptive capacity. We expect these firms to have larger productivity gains for two reasons: first, they have better ability to identify valuable and important technological developments in the foreign markets. When potential catch-up opportunities emerge, firms with no absorptive capacity may not be aware of them(Cohen and Levinthal, 1990). Second, they will be more efficient learners of foreign advanced technologies. Because of the accumulative nature of knowledge, firms equipped with more knowledge stock in a given filed may find it easier to catch-up with the recent technological developments in that field and other related fields（Grossman and Helpman,1993）. We estimate the instantaneous and long-run productivity effects of exporting using the data of China Annual Survey of Manufacturing Firms over the years 2001 -2007. Propensity score matching is used to control for selection into export. The main finding of this paper is that huge heterogeneity of productivity gains exists between firms with different pre-export 4

R&D status: For firms who have ever invested in R&D before export, starting to export raises a firm’s productivity by 16% instantaneously and 20% three years after starting to export. However, for firms who have not invested in R&D before export, starting to export basically has no significant effect in raising productivity. Moreover, when we compare the productivity gains from exporting for firms with different years of pre-export R&D investment, we find the gains are generally increasing in the number of years of pre-export R&D: firms with just 1 year of pre-export R&D have an instantaneous productivity gain around 8% , while firms with 3 years of pre-export R&D experience have an instantaneous productivity gain of 32%. These results suggest that firms with intentional and persistent investment in pre-export R&D are equipped with better absorptive capacity, and will therefore benefit from larger productivity gains after exporting. This paper is related to two strands of literature. The first strand is about testing the ―learning by export‖ hypothesis. So far there is no consensus on whether firms experience productivity gains after exporting, and the factors that affect the gains are still unclear. Studies on the export-productivity relationship using Chinese firm-level micro data has only recently begun (Lu et al., 2010 ; Park et al., 2010 ; Li and Yu, 2010). Some results are suggestive of productivity gains from exporting. For example, Park et al.(2010) find that firms whose export destinations experience large currency depreciations have slower growth in exports, and exports growth increase productivity. In a recent paper, Yang and Mallick (2010) find supportive evidence of learning by exporting using a small sample of Chinese firms over 2000 and 2002. Our paper is the first to investigate the productivity gains of exporting using the firm census data from the National Bureau of Statistics, which has wider coverage and is more 5

reliable. We have also found great heterogeneity of productivity gains by R&D investment status, which is absent in the previous studies. This paper is also related to the literature which studies the interaction between firms’ initial productivity, exporting behavior and productivity enhancement activities like R&D. Bustos(2010) finds that Argentinian firms in industries facing higher reductions in Brazil’s tariff increase their investment in technology faster, and the effect of tariffs on entry in the export market and technology upgrading is higher in the upper-middle range of firm size distribution. Lileeva and Trefler(2010) find there were labor productivity gains from exporting for low-productivity Canadian manufacturing plants that are induced to export because of the Canada-U.S. Free Trade Agreement, and the firms who gained did so by investing in technology. In these papers, technology enhancement activities like R&D have a direct effect in raising productivity i.e. the innovative effect mentioned in Cohen and Levinthal(1989), but the effect of increasing absorptive capacity is not mentioned. Aw,et al.(2007) is one of the few papers in this literature which mentioned the role of R&D in increasing absorptive capacity. Using Taiwanese electronics firm data during 1986 to 1996, they find supportive evidence of the interaction effects of R&D and exporting. However, in their later work using Taiwanese electronics firm data during 2000 and 2004, the interaction effects turn out to be negative. Therefore, that whether R&D investments increase the gains from exporting is still unclear and merits further analysis. Our paper provides new evidence for this question. The results found in this paper have important policy implications. Many developing countries have resorted to trade openness to boost economic growth. Although it is quite established that trade openness raises industry productivity at the aggregate level, more often 6

than not, the industry productivity growth is contributed by cross-firm reallocation of resources rather than with-in firm productivity growth(Bernard and Jenson,2004). Our results suggest that one explanation for the lack of within-firm productivity growth is the lack of absorptive capacity as a result of inadequate firm R&D. Therefore, policies that encourage firm R&D and other absorptive-capacity-building activities could be combined with trade policies in order to reap the full growth benefit of openness. The rest of the paper is organized as follows. Section 2 describes the data and conducts some preliminary analysis. Section 3 estimates the productivity gains from exporting using propensity score matching method. Section 4 evaluates the role of pre-export R&D on post-entry productivity gains. Section 5 checks for robustness and makes some further discussion. The last section concludes.

2.

Data description The data used in this paper comes from a rich firm-level panel dataset collected and maintained by the National Bureau o f Statistics in an annual survey of manufacturing firms. The dataset covers all SOEs and non-SOEs that are ―above scale‖,i.e. with annual sales above 5 million yuan during 2001 and 2007. On average, more than 200 thousand firms are included each year, covering 33 industries and 31 provinces. To clean the data, we drop the observations if (1) the observation reports missing or negative value on any of the following variables: overall revenue, total employment, fixed capital, total sales, intermediate inputs, R&D expenditures , export value. (2) The observation has employment less than 8 (3) The observation’s value of total sales is small than export value. The final sample we use for the 7

subsequent analysis contains 490,302 firms with 1,592,246 observations. Table 1 summarizes the exporting and R&D status of the sampled Chinese firms. On average, about 27% firms exported and about 12% firms conducted R&D each year during 2001 and 2007. The fact that the some exporters do not invest in R&D is consistent with Bustos(2010), and this provides us with a variation of the data that can help us exploit the role of R&D in generating post-entry productivity gains, as we will show in Section 4. We further divide the firm’s exporting status into three categories: New-exporters, existing exporters and never-exporters. New exporters are the firms whose year of first-time export is later than the year they’re for the first time observed in the sample; Existing exporters are the firms that already export when they’re observed in the data for the first time;5 and never exporters are the firms who don’t export during the whole sample period. In the sample, new exporters constitute about 27% of all firms and never exporters constitute another 63%, with the rest being existing exporters.6 Since our main interest is the productivity gains when and after a firm starts to export, comparing the post-entry productivity of new exporters and never exporters will be the focus of our following analysis.

5

Since we don’t know whether a firm exports in the periods before it is observed, we assume all firms that export

when they’re for the first time observed have already begun exporting before they’re observed. 6

Note that new exporter and existing exporter altogether accounts for 37% of all firms, which is higher than the

share of exporter in Table 1 (27.1%).The reason is that the exporter share in Table 1 is calculated by averaging the share of exporter in each year, while the status of new exporter and existing exporter doesn’t depend on year. For example, If a firm first appears in the sample in 2001, start exporting in 2003 and stops exporting in 2005, it will not be included in the exporter share in 2006 and 2007, but it will still be included as a ―new exporter‖. In other words, the share of new exporter and existing exporter added together equals the share of firms that have ever exported during the sample period.

8

[Insert Table 1 Here] Table 2 summarizes firm’s R&D pattern by their exporting status, firm size and ownership structure.7 It is easily seen that log R&D expenditure and the share of firms that conduct R&D are higher among exporter, large sized firms and foreign owned firms. It is important to bear in mind that R&D expenditures are correlated with these firm characteristics. So we need to control for these characteristics in our estimation when we identify the effect of R&D on the productivity gains from exporting. We return to this point in section 5. [Insert Table 2 Here]

3. Estimating the post-entry productivity gain 3.1 Estimating the total factor productivity As productivity evolution is the focus of our analysis, it is crucial to get a precise measure of firms’ productivity. Following the standard literature, we use total factor productivity (TFP) as the measure of productivity. We calculate TFP using the method proposed by Olley and Pakes(1996), which uses firms’ investment and capital stock as proxies for the unobservable productivity. The strength of this estimation procedure lies in two aspects: (1) it allows for controlling the simultaneity bias when estimating the production function, without relying on instruments which are often unobtainable.(2) it corrects for the sample selection bias in the panel data, i.e. only more productive firms are able to survive and remain observable in the sample. The detailed estimating 7

Firm size is proxied by firm’s fixed capital stock as in Lu et al.(2010). We discretize size by dividing firms into

small and large sized firms, with small sized firms referring to firms whose capital stock are below the median of the industry-year cell they belong to, and large firms those above the median.

9

procedure of productivity using the OP method is described in Appendix A. Figure 1 plots the TFP evolution for exporters and non-exporters, respectively. Two patterns emerge immediately: first, like in almost any other country, exporters are more productive. In the figure, exporters generally have a 5% productivity premium over non-exporters. Whether this productivity premium comes from selection into export or learning by exporting is discussed in the next subsection. Second, the evolution of productivity is quite similar between exporters and non-exporters, reflecting the common macro economic factors that are affecting the productivity of all firms. Note that TFP experienced a dip in both 2003 and 2004. One possible explanation for this is the SARS crisis that swept China throughout 2003 and the ensuing high inflation in 2004. [Insert Figure 1 Here]

3.2 Estimating the post-entry productivity effect We now turn to econometrics to rigorously examine the post-entry productivity effect of exporting. The basic idea is to see exporting as a ―treatment‖, so its effects can be evaluated by the standard methods of evaluating the Average Treatment Effect on the Treated(ATT). Denote the outcome variable for firm i at time t as

it1 if it starts to export at time k , and it 0 if it

doesn’t. The average treatment effect of start to export, on the starting firms, can be represented as

E(it1  it 0 | Startik  1)  E(it1 | Startik  1)  E(it 0 | Startik  1) Where Startik  1 if the firm i starts to export at time k . The practical difficulty is that

(1)

it 0 is

not observable. In order to get unbiased estimates of the ATT, we need to construct a control group consisting of firms that are conceptually identical to the starters had they not start to export. Following the recent literature (Wagner,2002; Greenaway and Kneller, 2008; Alvarez and 10

López,2005; De Loecker, 2007), we adopt the propensity score matching method to construct such control groups. We use the information prior to the year of exporting to estimate the propensity score:

P(Startit )  (h( Xit  1))

(2)

Xit  1 includes a series of firm characteristics at t  1 which predicts whether or not a firm exports at time t . As suggested by the literature, the most important characteristic is productivity. Other variables include firm size as measured by total employment, fixed capital, and a set of region, ownership and industry dummies. To get an unbiased estimation of the propensity score, we allow for a flexible functional form of h(.) by including higher order and interaction terms. The propensity score is estimated year-by-year. We conduct the propensity score matching following the algorithm in Greenaway and Kneller(2008) and De Loecker(2007).9 To avoid the risk of comparing firms that are affected by different macroeconomic and industry conditions, the matching is conducted on a year-by-year, industry-by-industry basis. After identifying the matched pairs, we pool all the years and industries together and calculate the average difference between the treated and the control group. Specifically, we calculate the following matching estimator:

ATTs 

Where 9

1 NT

2007

  (

k  2002 g

1 igk  s

  c igk  s )

s  0 , 1, 2 , . . .

(3)

i

1igk  s denotes the productivity of the treated firm i from industry g at time k  s .

The exact procedure of matching goes as follows: (1) divide the propensity score into

k

equally spaced

intervals so that the average propensity scores of the treated and control group do not differ. (2) within each interval, test whether the first moment of the covariates differ between the treatment and control groups. i.e. to test the balancing condition. (3) if the balancing condition is rejected, alter the functional form of the propensity score by further adding higher order and interaction terms, and repeat the procedure of (1)-(2).(4) After the balancing condition is satisfied, match the sample based on nearest neighbors.

11

Here k represents the year firm i starts to export and s denotes the number of years after firm i starts to export.

 c igk  s denotes the productivity for the matched firm for firm i in the

year k  s . N T denotes the number of treated firms, i.e. the total number of new exporters.10 It should be noted that for the long-run effects of exporting, we only estimated the s year effects for the firm that continues to export in year t  s . Therefore the long-run effect doesn’t apply to those firms that exit export market after the first year. Since it is not unusual for firms to exit the export market after their first year of entry (Eaton et al.,2008), we expect the number of treated units to shrink by a large extent when estimating the long-run effects. Table 3 reports the matching results. In the first part, the productivity level is used as the outcome variable. The results show that on average, the sampled Chinese manufacturing firms experience only weak gains from exporting over the years 2001-2007: productivity gains from exporting is significant only in the first year they start to export. Since the TFP is estimated after taking logs, the estimated ATT has a percentage change interpretation: Controlling for selection into export, the exporting firms are 2% more productive than the matched non-exporters at the year when they start to export (instantaneous effect). While after three years, the productivity gains from exporting (long-term effect) no longer exist. This result is consistent with the general pattern found in other previous studies that productivity gains are largest in the first year that firms start exporting (Damijan and Kostevc,2005; Greenaway and Kneller,2008; Martins et al., 2009). This finding might be subject to several interpretations. First, firms may learn most in their initial 10

The standard error of the estimator is calculated by the square root of the following formula

w

i

2

1 Var (1 )  i 2 Var ( c ) ,where N T and N C are the number of treated and control units, NT NC

1 and  c the outcome of treated and control units, and wi is the weight assigned to each control units in the nearest neighbor matching. IC is the set of control units. respectively;

12

period of globalization, when they’re for the first time exposed to advanced foreign technologies and faced with foreign competition. As their exporting experiences grow, there will be less to learn from the foreign market. Second, the increase of productivity in the first year of exporting may simply reflect greater utilization of firms’ production capacity after suddenly getting access to the foreign demand(Damijan and Kostevc,2005). In the second part, year-on-year productivity growth rate is used as the outcome. The results show that in most of the years we see insignificant effect, but in the third year after exporting the effect becomes significant, though the magnitude is still negligibly small. [Insert Table 3 Here]

4. The role of pre-export R&D in generating post-entry productivity gains

In this section we argue that pre-export R&D is important in generating post-entry productivity gains. As described in the introduction, R&D not only has direct effects on productivity by improving production process or innovation of higher quality goods, it also enriches the firm’s knowledge stock, increases its ability to identify and absorb advanced technology and thus equips the firm to exploit future productivity enhancement opportunities. Following this logic, when exposed to the foreign technologies after exporting, firms that have higher absorptive capacity via investing in pre-export R&D will learn more from exporting and therefore experience more productivity gains. In the following subsections we test this hypothesis.

4.1. Productivity effects of exporting for firms with/without Pre-export R&D

We firstly divide the sample to two sub-groups: firms who have conducted any R&D before 13

exporting and those who have not. Specifically, we generate a dummy variable ―Pre-exR&D‖, which is equal to 1 if the R&D expenditure of the firm is greater than zero in at least one year before it starts to export.11 We divide the sample according to the value of this dummy and then perform propensity score matching to each subsample separately. It should be noted that for the first subsample, our matching approach is actually comparing the outcome between the firms who both export and conduct pre-export R&D (treatment group) and non-exporters(control group). Similarly, for the second subsample, we are comparing exporters with no pre-export R&D and non-exporters. If pre-export R&D does increase the firm’s absorptive capacity and increase the extent of post-entry productivity gain, we will see greater gains from exporting in the group with pre-export R&D. The results of the two subsamples are shown in Table 4.ATTs for the subsample with pre-export R&D are shown in Panel (A) while ATTs for the subsample without pre-export R&D are shown in Panel (B).The results show that productivity gains from exporting have huge difference between subsamples: In Panel(A), all of the ATTs-from the first year of export to three years after starting to export- are positively significant, and the magnitude stays around 14%-20%, which is much higher than the 2% effect we obtained using the full sample in section 3. In Panel (B), however, none of the ATTs are positively significant. These results are strongly supportive of the hypothesis that pre-export R&D helps generate greater productivity gains from exporting: while the productivity gains from exporting are weak and transient for all firms on average(as

11

Ideally, pre-export R&D expenditure should be treated as a continuous variable. However, treating the

pre-export R&D status as a dummy variable has the advantage of allowing us to perform the matching technique throughout the paper, so that results obtained in this section can be compared with the results in section 3.

14

shown in section 3), the gains are large and lasting for firms with pre-export R&D. The effect is large because firms with more absorptive capacity might be more able to recognize the most effective productivity-enhancing technology and learn them more efficiently. The effect is lasting because firms with more absorptive capacity might be able to continuously discover new learning opportunities in their export process. For firms without pre-export R&D, however, productivity gains do not exist even instantaneously. [Insert Table 4 Here]

4.2 Productivity effects for firms with different years of pre-exporting experience In section 4.1 we compared the productivity effects of exporting for firms with and without pre-export R&D. This method provides us with a stark demonstration that pre-export R&D helps to generate large and lasting productivity gains from exporting. However, the 0-1 classification of R&D status doesn’t distinguish the firms that have intentional and persistent R&D investment from the firms that are only accidentally involved in R&D activities. As pointed out by Cohen and Levinthal (1989), absorptive capacity is built by intentional and persistent R&D investment. Therefore, we may expect a firm’s absorptive capacity, and thus its productivity gains from exporting, to depend on the extent of its pre-export R&D investment. The firms that persistently invest in R&D activities before exporting will have more absorptive capacity, and thus experience larger productivity gains from exporting, than the firms that are accidentally involved in R&D activities. One natural proxy for the extent of the pre-export R&D is the number of years a firm has

15

invested in pre-export R&D.12 Therefore, we divide the firms into several subsamples by this variable, and conduct the matching exercises as before for each subsample. The results are shown in Table 5. Generally, we find the productivity effects of exporting to be increasing in the number of years a firm has conducted pre-export R&D: For firms with one year of pre-export R&D, the instantaneous effect and the two-year effect are 8% and 20%, respectively, which are smaller than the 16% and 22% effect found in section 4.1 for all firms with pre-export R&D .The two effects are much higher for firms with three years pre-export R&D: the instantaneous effect is 32% and the two-year effect is 34%.13 These results are supportive of our hypothesis that firms with intentional and persistent pre-export R&D will have better absorptive capacity, and will benefit from greater productivity gains from exporting. [Insert Table 5 Here]

5. Robustness and further discussion 5.1.Alternative Measures of productivity In order to check whether our main results are sensitive to the estimation method of productivity, we also estimated TFP using the methods suggested by Levinsohn and Petrin(2003), which uses the intermediate inputs to proxy for the unobservable productivity.14 Table 6 reports

12

This proxy is suggested in an influential paper by Zahra and George(2002)

13

The instantaneous effects for firms with 5 or 6 years of pre-export R&D are even larger, but they’re not

statistically significant. This may due to too few matched observations in these categories. 14

Compared with Olley and Pakes(1996), the method of Levinsohn and Petrin(2003) has the advantage that it

only requires the data of intermediate inputs to proxy for productivity, instead of requiring the investments data which in many cases are reported to be zero or missing.

16

the summary statistics of the two TFP measures. [Insert Table 6 Here] We firstly replicated our matching exercise on the subsamples with and without pre-export R&D, as in section 4.1, but now use the TFP calculated by LP method instead of OP. The results are reported in Table 7. Qualitatively, our main argument is again confirmed: firms with pre-export R&D experience larger and more lasting productivity gains from exporting: All ATTs for the subsample with pre-export R&D are positively significant, with the instantaneous effect being 20% and the three-year effect being 26%. For the subsample with no pre-export R&D, the productivity effects estimated using TFP LP are all positive, but the instantaneous and three-year effects are insignificant. The one-year and two-year effects, though significant, are much smaller than the corresponding effects found in subsample with pre-export R&D. [Insert Table 7 Here] We also replicated the matching exercise for subsamples with different number of years of pre-export R&D, as in section 4.2. The results in Table 8 confirms that firms’ productivity gains from exporting increase with the number of years for which they invested in pre-export R&D: the instantaneous effect and two year effects are 15% and 20% for firms with one year pre-export R&D but 29% and 47% for firms with three years pre-export R&D. In addition, the instantaneous effect for firms with 4 years of pre-export R&D now becomes significant, because of more matched observations. [Insert Table 8 Here]

5.2 Controlling for other confounding variables 17

One might worry about the possibilities that the huge productivity gains heterogeneity found in the previous section is actually picking up the effect of other variables that are correlated with firm R&D instead of caused by R&D itself. One possibility is firm size, as we showed in section 2, R&D investment is highly concentrated in large firms. If large firms systematically gain more from exporting, the result in the previous section is likely to emerge even if R&D doesn’t matter. The other possibility is firm’s ownership. Firms that have foreign backgrounds may be more efficient in absorbing external knowledge and more responsive to the advanced foreign technology. Since R&D are also higher in foreign owned firms, the result in the previous section may have just picked up the effect of foreign ownership. To make sure that it is R&D that matters, we need to control for these confounding variables. One method is to divide the firms into subsamples by these confounding variables, and do the matching by pre-export R&D status in each subsample to see if the main conclusion of section 4 holds irrespective of the confounding variables. In Table 9, we report the matching results on subsamples of small and large firms, where small and large firms are as defined in section 2. We would like to compare the productivity gains between firms with and without pre-export R&D within each firm size category. Firstly, within small-sized firms, exporters that have pre-export R&D have an instantaneous productivity gain of 9% and a two year gain of 5%. However, for firms without R&D, neither the instantaneous nor two year effect is positive.15 Secondly, for large sized firms, the instantaneous and two year effect are 18% and 22% for firms with pre-export R&D, while for firms without pre-export R&D, neither effect is positively significant. Therefore,

15

Although the two year effect for small sized firms with pre-export R&D is insignificant due to too few matched

firms, the point estimate is still much larger than the point estimate for firms without pre-export R&D.

18

our main result in section 4 still holds irrespective of the size of the firm. [Insert Table 9 Here] In Table 10 we report matching results on foreign owned firms and domestic firms, respectively. Again, our main result is not sensitive to the ownership of the firm. Within foreign owned firms, the instantaneous and two year effect for firms with pre-export R&D are 28% and 22%, respectively, while for firms without pre-export R&D neither effect is positively significant. Similar pattern holds for domestically owned firms. [Insert Table 10 Here]

5.3

Pre-export R&D and productivity gains: industry heterogeneity We have seen in section 4 that pre-export R&D has a significant impact on productivity gains

from exporting. However, does this impact differ across industries? Intuition tells us absorptive capacity matters more for skill-intensive industries, so we should expect the impact of pre-export R&D to be larger in skill-intensive industries. To test this, we calculate the difference between the productivity gains for firms with pre-export R&D and for firms without pre-export R&D, within each 2-digit industry. We then plot in Figure 2 the difference against each industry’s R&D intensity, which we use as a proxy for skill intensity.

16

We expect to observe a positive relation

because we expect the impact of pre-export R&D to be larger in skill-intensive industries. The pattern emerged in Figure 2 turns out to be weakly supportive of this hypothesis. A clear positive relation is detected in the lower-left part of the graph, though there exist several outliers industries. [Insert Figure 2 Here] 16

R&D intensity is calculated by R&D expenditure divided by overall revenue

19

6. Conclusion While it has been well established in the recent firm-level trade literature that exporters are productive, the evidence is mixed concerning whether export has generated productivity gains for firms. Unveiling the variables that are affecting the productivity gains from exporting might be an important step in understanding and unifying the mixed evidence. In this paper, we argue that pre-export R&D is a crucial factor for generating productivity gains from exporting. R&D before export increase firms’ absorptive capacity and enhances firms’ learning efficiency when exposed to advanced foreign technology. To test this hypothesis, we estimate the instantaneous and long-run productivity effects of starting to export on the universe of Chinese manufacturing firms over the years 2001-2007. Propensity score matching is used to control for selection into export. Our results show that (1) for all sampled firms on average, starting to export has an instantaneous effect in raising productivity by 2%, but there are no significant long-run effects. (2) While the productivity gains from exporting are weak and transient for all firms on average, the gains are large and lasting for firms with pre-export R&D, with the instantaneous and three-year effect being 16% and 20%, respectively. For firms without pre-export R&D, however, exporting has no significant productivity effects even instantaneously. (3) The productivity gains from exporting are increasing in the number of years a firm invests in pre-export R&D: for firms with one year export R&D, the instantaneous and two-year effects are 8% and 20%, respectively. While for firms with three years of pre-export R&D, these two effects are 32% and 34%, suggesting that firms involved in intentional and persistent R&D activities benefit from more productivity gains from exporting. Our qualitative results are robust to alternative TFP measures and 20

controlling for other confounding variables. We believe these results provide strong evidence that absorptive capacity built by pre-export R&D does help firms learn from exporting and benefit from productivity gains.

21

Reference [1] Alvarez ,Roberto & López,, Ricardo 2005. "Exporting and performance: evidence from Chilean plants," Canadian Journal of Economics, Canadian Economics Association, vol. 38(4), pages 1384-1400, November. [2] Arnold, Jens Matthias & Hussinger, Katrin, 2005. "Export Behavior and Firm Productivity in German Manufacturing: A Firm-Level Analysis," Review of World Economics (Weltwirtschaftliches Archiv), Springer, vol. 141(2), pages 219-243, July [3] Aw, Bee Yan & Roberts, Mark J. & Winston, Tor 2007. "Export Market Participation, Investments in R&D and Worker Training, and the Evolution of Firm Productivity," The World Economy, Blackwell Publishing, vol. 30(1), pages 83-104, 01. [4] Aw, Bee Yan & Roberts, Mark J. & Xu, Daniel Yi 2008. "R&D Investments, Exporting, and the Evolution of Firm Productivity," American Economic Review, American Economic Association, vol. 98(2), pages 451-56, May. [5] Bernard, Andrew B. & Bradford Jensen, J., 1999. "Exceptional exporter performance: cause, effect, or both?," Journal of International Economics, Elsevier, vol. 47(1), pages 1-25, February. [6] Bernard, Andrew B. & Bradford Jensen, J. , "Exporting and Productivity in the USA," Oxford Review of Economic Policy, Oxford University Press, vol. 20(3), pages 343-357, Autumn. [7] Bustos, Paula ,2010. "Trade Liberalization, Exports and Technology Upgrading: Evidence on the Impact of MERCOSUR on Argentinean Firms," American Economic Review, forthcoming. [8] Clerides, Sofronis K.& Saul Lach & Tybout, James R. 1998. "Is Learning By Exporting Important? Micro-Dynamic Evidence From Colombia, Mexico, And Morocco," The Quarterly Journal of Economics, MIT Press, vol. 113(3), pages 903-947, August. [9] Cohen, Wesley M & Levinthal, Daniel A, 1989. "Innovation and Learning: The Two Faces of R&D," Economic Journal, Royal Economic Society, vol. 99(397), pages 569-96, September [10] Cohen, Wesley M.& Levinthal Daniel A. 1990. "Absorptive Capacity: A New Perspective on Learning and Innovation." Administrative Science Quarterly Vol. 35, No. 1, Special Issue: Technology, Organizations, and Innovation, pp. 128-152 [11] Damijan,Joze P. & Kostevc,Crt, 2005. "Performance on Exports: Continuous Productivity Improvements or Capacity Utilization," LICOS Discussion Papers 16305, LICOS - Centre for Institutions and Economic Performance, K.U.Leuven. [12] De Loecker, Jan, 2007. "Do exports generate higher productivity? Evidence from Slovenia." Journal of International Economics, Elsevier, vol. 73(1), pages 69-98, September. 22

[13] De Loecker, Jan,2010. ―A Note on Detecting Learning by Exporting, NBER working paper.‖ [14] Eaton, Jonathan, Marcela Eslava, Maurice Kugler, and James Tybout (2008) ―Export Dynamics in Colombia: Firm Level Evidence,‖ in The Organization of Firms in a Global Economy, edited by Elhanan Helpman, Dalia Marin, and Thierry Verdier. Cambridge: Harvard University Press,2008. [15] Girma, S. ,2002: "Absorptive capacity and productivity spillovers from FDI: A threshold regression analysis," Research Paper 2002/08, Leverhulme Centre for Research on Globalisation and Economic Policy, University of Nottingham. [16] Girma, S., Greenaway, A. and Kneller, R.,2004, "Does Exporting Increase Productivity? A Microeconometric Analysis of Matched Firms." Review of International Economics, 12: 855–866. [17] Greenaway, David & Yu, Zhihong 2004. "Firm-level interactions between exporting and productivity: Industry-specific evidence," Review of World Economics (Weltwirtschaftliches Archiv), Springer, vol. 140(3), pages 376-392, September. [18] Greenaway, David & Gullstrand ,Joakim & Kneller,Richard, 2005. "Exporting May Not Always Boost Firm Productivity," Review of World Economics (Weltwirtschaftliches Archiv), Springer, vol. 141(4), pages 561-582, December. [19] Greenaway, David & Kneller, Richard, 2007. "Industry Differences in the Effect of Export Market Entry: Learning by Exporting?," Review of World Economics (Weltwirtschaftliches Archiv), Springer, vol. 143(3), pages 416-432, October. [20] Greenaway, David & Kneller, Richard, 2008. "Exporting, productivity and agglomeration," European Economic Review, Elsevier, vol. 52(5), pages 919-939, July. [21] Griffith, Rachel & Redding, Stephen & Reenen, J. Van, 2004. "Mapping the Two Faces of R&D: Productivity Growth in a Panel of OECD Industries," The Review of Economics and Statistics, MIT Press, vol. 86(4), pages 883-895, December. [22] Grossman, Gene M. & Helpman, Elhanan, 1993. "Innovation and Growth in the Global Economy," MIT Press Books,The MIT Press,edition 1, volume 1, number 0262570971, June. [23] Hahn,Chin Hee,2004."Exporting and Performance of Plants: Evidence from Korean Manufacturing," NBER Working Papers 10208, National Bureau of Economic Research, Inc. [24] Hu ,Albert G. Z. & Gary H. Jefferson & Qian Jinchang, 2005. "R&D and Technology Transfer: Firm-Level Evidence from Chinese Industry," The Review of Economics and Statistics, MIT Press, vol. 87(4), pages 780-786, 01.

23

[25] Imbens and Woodridge, 2007. ―Estimation of Average Treatment Effects Under Unconfoundedness‖, ―What’s New in Econometrics‖ Lecture Notes series [26] Kinoshita, Yuko, 2001. "R&D and Technology Spillovers through FDI: Innovation and Absorptive Capacity," CEPR Discussion Papers 2775, C.E.P.R. Discussion Papers. [27] Levinsohn,James & Petrin, Amil, 2003. "Estimating Production Functions Using Inputs to Control for Unobservables," Review of Economic Studies, Blackwell Publishing, vol. 70(2), pages 317-341, 04. [28] Feenstra, Robert, Li, Zhiyuan & Yu, Miaojie, 2010. "Exports and Credit Constraint under Private Information: Theory and Appliaction to China," University of California, Davis. [29] Lileeva,Alla & Trefler, Daniel, 2010. "Improved Access to Foreign Markets Raises Plant-Level Productivity... for Some Plants," The Quarterly Journal of Economics, MIT Press, vol. 125(3), pages 1051-1099, August. [30] Lu, Jiangyong & Lu, Yi & Tao, Zhigang, 2010. "Exporting behavior of foreign affiliates: Theory and evidence," Journal of International Economics, Elsevier, vol. 81(2), pages 197-205, July. [31] Martins, Pedro & Yang,Yong, 2009. "The impact of exporting on firm productivity: a meta-analysis of the learning-by-exporting hypothesis." Review of World Economics, Springer, vol. 145(3), pages 431-445, October. [32] Melitz, Marc J., 2003. "The Impact of Trade on Intra-Industry Reallocations and Aggregate Industry Productivity," Econometrica, Econometric Society, vol. 71(6), pages 1695-1725, November. [33] Olley, G Steven & Pakes, Ariel, 1996. "The Dynamics of Productivity in the Telecommunications Equipment Industry," Econometrica, Econometric Society, vol. 64(6), pages 1263-97, November. [34] Park,Albert & Dean Yang & Shi ,Xinzheng & Jiang,Yuan 2006. "Exporting and Firm Performance: Chinese Exporters and the Asian Financial Crisis," Working Papers 549, Research Seminar in International Economics, University of Michigan. [35] Wagner, Joachim, 2002. "The causal effects of exports on firm size and labor productivity: first evidence from a matching approach," Economics Letters, Elsevier, vol. 77(2), pages 287-292, October. [36] Woodridge, 2002. J. Woodridge, Econometric analysis of cross section and panel data. , The MIT Press, Cambridge, MA (2002).

24

[37] Yang, Yong and Mallick, Sushanta 2010 , ―Export Premium,Self-selection, and Learning-by-exporting: Evidence from Matched Chinese Firms‖, The World Economy, Volume 33, Number 10, October 2010 , pp. 1218-1240(23) [38] Zahra, Shaker A. and George,Gerard,2002."Absorptive Capacity: A Review, Reconceptualization, and Extension." The Academy of Management Review, Vol. 27, No. 2. (2002), pp. 185-203.

25

Table 1 Exporting and R&D status of Chinese manufacturing firms

Variable Exporter

Annual Average(2001-2007) 27.14%

R&D

12.01%

New exporter

27.57 %

Existing exporter

8.83%

Never exporter Num.of Obs.

63.58% 1,592,246

Note: ―Exporter‖and ―R&D‖ represent the year average of the share of exporters and the share of firms which have positive R&D expenditures, respectively. ―New exporter‖ represents the share of firms whose year of first-time export is later than the year they’re for the first time observed in the sample; ―Existing exporter‖ represents the share of firms that are the firms that already export when they’re observed in the data for the first time. ―Never exporter‖ represents the share of firms that do not export during the whole sample period.

Table 2 R&D expenditure and share of firms with R&D by various firm characteristics Exporter

Non-exporter

Small size

Large size

FIE

Non-FIE

Log R&D expenditure

0.92

0.48

0.30

0.92

0.66

0.59

Share of firms with R&D

0.16

0.10

0.07

0.16

0.12

0.11

26

Table 3 :Productivity gains from exporting (full sample) s

0

1

2

3

ATTs

0.0210*

0.0111

0.0230

0.0341

Std. Err.

(0.0122)

(0.0223)

(0.0302)

(0.0398)

17357

8371

4291

2306

0.0023

0.0106

0.0075*

0.0056

(0.0063)

(0.0101)

(0.0040)

(0.0079)

12374

5787

3218

1798

(1) Productivity

Nr of Treated Units (2) Growth rate ATTs Std. Err. Nr of Treated Units

Note: This table demonstrates the estimated average gains from exporting using the matching approach. s indicates the number of years after the firm starts to export for the first time. Results using productivity level are reported in (1) and those using growth rate are reported in (2). *,** and *** indicates significance at 10%, 5% and 1% level,respectively.

Table 4: Productivity gains from exporting(for subsamples with and without pre-export R&D) s

0

1

2

3

0.1644***

0.1377***

0.2163***

0.1956***

(0.0295)

(0.0417)

(0.0570)

(0.0757)

3421

1715

958

548

Panel (A): Results for firms with pre-export R&D ATTs Std. Err. Nr of Treated Units.

Panel (B): Results for firms without pre-export R&D ATTs

-0.0147

-0.0216

-0.0325

-0.0161

Std. Err.

(0.0134)

(0.0193)

(0.0275)

(0.0366)

13936

6656

3333

1758

Nr of Treated Units.

Note: This table demonstrates the estimated average gains from exporting using the matching approach. s indicates the number of years after the firm starts to export for the first time. Panel (A) shows the result using only the subsample of firms which have invested in R&D before export. Panel (B) shows the results using only the subsample of firms that have not invested in pre-export R&D. The outcome variable is productivity level. *,** and *** indicates significance at 10%, 5% and 1% level, respectively.

27

Table 5: Productivity gains from exporting for subsamples with different years of pre-export R&D s

0

1

2

3

4

5

6

Panel(A): Instantaneous effect ATTs

-0.0147

0.0754*

0.2336***

0.3188***

0.2222*

0.3644

0.3572

Std. Err.

(0.0134)

(0.0466)

(0.0729)

(0.1025)

(0.1518)

(0.2743)

(0.4262)

13936

1809

842

480

205

61

24

Nr of T.U.

Panel(B): Two-year effect ATTs

-0.0325

0.1997***

0.2447***

0.3350***

0.1204

n.a.

n.a.

Std. Err.

(0.0275)

(0.0733)

(0.1210)

(0.1480)

(0.2911)

n.a.

n.a.

3333

615

202

114

27

n.a.

n.a.

Nr of T.U.

Note: s indicates the number of years a firm conducts pre-export R&D investment. Panel (A) shows the instantaneous effect, using the matching approach in section3; Panel (B) shows the two-year effect. The outcome variables is the productivity level. *,** and *** indicates significance at 10%, 5% and 1% level, respectively.

Table 6: Summary statistics for TFP_OP and TFP_LP Variable

Mean

Std. Dev.

Min

Max

TFP_OP

4.214

1.151

-8.410

10.590

TFP_LP

3.433

1.891

-7.375

14.651

28

Table 7: Productivity gains from exporting for subsamples with and without pre-export R&D (Using TFP_LP) s

0

1

2

3

0.1997***

0.1777***

0.2578***

0.2630***

(0.0348)

(0.0501)

(0.0647)

(0.0832)

5996

3079

1749

1059

0.0154

0.0436*

0.0609*

0.0597

(0.0170)

(0.0250)

(0.0342)

(0.0453)

26226

13003

6557

3867

Panel (A): Results for firms with pre-export R&D (1) Productivity ATTs Std. Err. Nr of Treated Units.

Panel (B): Results for firms without pre-export R&D (1) Productivity ATTs Std. Err. Nr of Treated Units.

Note: This table demonstrates the estimated average gains from exporting using the matching approach and TFP_LP as the outcome variable. s indicates the number of years after the firm starts to export for the first time. Panel (A) shows the result using only the subsample of firms that have invested in R&D before export. Panel (B) shows the results using only the subsample of firms that have not invested in pre-export R&D. *,** and *** indicates significance at 10%, 5% and 1% level, respectively.

Table 8: Productivity gains from exporting for subsamples with different years of pre-export R&D (Using TFP_LP) s

0

1

2

3

4

5

6

Panel(A): Instantaneous effect ATTs Std. Err. Nr of T.U.

0.0154

0.1502***

0.2255***

0.2868***

0.3687***

0.3511

0.2371

(0.0170)

(0.0466)

(0.0729)

(0.1025)

(0.1518)

(0.2743)

(0.4262)

26226

3426

1399

708

323

99

41

Panel(B): Two-year effect ATTs

0.0609*

0.2028***

0.3205***

0.4715***

0.3908

n.a.

n.a.

Std. Err.

(0.0342)

(0.0797)

(0.1430)

(0.2124)

(0.3939)

n.a.

n.a.

6557

1172

364

165

48

n.a.

n.a.

Nr of T.U.

Note: s indicates the number of years a firm conducts pre-export R&D investment. Panel (A) shows the instantaneous effect, using the matching approach in section3; Panel (B) shows the two-year effect. The outcome variables is the productivity level. *,** and *** indicates significance at 10%, 5% and 1% level, respectively.

29

Table 9 : Instantaneous and two-year productivity effects of exporting, by pre-export R&D status within firm size. Instantaneous effect Firm size

small

Two year effect

large

small

large

Panel (A): Results for firms with pre-export R&D Coef.

0.0905*

0.1847***

0.0540

0.2188***

Std. Err.

(0.0538)

(0.0345)

(0.1374)

(0.0628)

844

2655

159

794

Nr of Treated Units.

Panel (B): Results for firms without pre-export R&D Coef.

-0.0056

-0.0004

-0.0590

-0.0007

Std. Err.

(0.0191)

(0.0188)

(0.0409)

(0.0363)

5939

7855

1304

1995

Nr of Treated Units.

Note: This table reports instantaneous and two-year productivity effect of exporting for firms with pre-export R&D or without pre-export R&D, conditioning on firm size(small/large). Panel A (Panel B) for results within the subsample with (without) pre-export R&D. Dependent Variable is productivity level. *,**,*** refers to significant at 10%,5%,1%, respectively.

Table 10 : Instantaneous and two-year productivity effects of exporting, by pre-export R&D status within ownership type. Instantaneous effect Ownership type

FIE

Two year effect

Non-FIE

FIE

Non-FIE

Panel (A): Results for firms with pre-export R&D Coef. Std. Err.

0.2797***

0.1578***

0.2203*

0.2197***

(0.0706)

(0.0324)

(0.1250)

(0.0641)

629

2765

214

732

Nr of Treated Units.

Panel (B): Results for firms without pre-export R&D Coef.

-0.0469

0.0029

-0.0692

-0.016

Std. Err.

(0.0285)

(0.0152)

(0.0495)

(0.0328)

3454

10379

1109

2204

Nr of Treated Units.

Note: This table reports instantaneous and two-year productivity effect of exporting for firms with pre-export R&D or without pre-export R&D, conditioning on whether a firm is FIE or not. Panel A (Panel B) for results within the subsample with (without) pre-export R&D. Dependent Variable is productivity level. *,**,*** refers to significant at 10%,5%,1%, respectively. 30

0

.5

1

1.5

Figure 2: ATT difference and industry R&D intensity

-.5

tfp gain difference

Figure 1: TFP evolution for Chinese manufacturing firms: 2001-2007

0

1

2 (R&D intensiy)*100

31

3

Appendix A: Estimate TFP by Olley-Pakes (1996)

Econometricians have tried hard to address these empirical challenges, but were unsuccessful until the pioneering work by Olley and Pakes (1996). In the beginning, researchers used two-way (i.e., firm-specific and year-specific) fixed effects estimations to mitigate simultaneity bias. Although the fixed effect approach controls for some unobserved productivity shocks, it does not offer much help in dealing with reverse endogeneity and remains unsatisfactory. Similarly, to mitigate selection bias, one might estimate a balanced panel by dropping those observations that disappeared during the period of investigation. The problem is that a substantial part of information contained in the dataset is wasted, and the firm's dynamic behavior is completely unknown. Fortunately, the Olley--Pakes methodology makes a significant contribution in addressing these two empirical challenges. Consider a standard Cobb-Douglus production function：

Yit= itMit  m Kit  k Lit  l

(1)

where Yit is the value-added production of firm i at year t , Kit , Lit and Mit denotes labor, capital, and intermediate inputs, respectively. By assuming that the expectation of future realization of the unobserved productivity shock, vit , relies on its contemporaneous value, the firm i 's investment is modeled as an increasing function of both unobserved productivity and log capital, kit  ln Kit .Following previous works, such as van Biesebroeck (2005) and Amiti and Konings (2007), the Olley--Pakes approach was revised by adding the firm's export decision as an extra argument of the investment function since most firms' export decisions are determined in the previous period (Tybout, 2003):

32

Iit  I (ln Kit , vit , EFit )

(2)

where EFit is a dummy to measure whether firm i exports in year t . Therefore, the inverse function of Iit is

vit  I 1 (ln Kit , Iit , EFit )

18

(3)

The unobserved productivity also depends on log capital and the firm's export decisions. Accordingly, the estimation specification can now be written as:

ln Yit   0   m ln Mit   l ln Lit  g (ln Kit, Iit, EFit )   it where g (ln Kit , Iit , EFit ) is defined as

(4)

 k ln Kit  I 1 (ln Kit , Iit, EFit ) . Following Olley and

Pakes (1996) and Amiti and Konings (2007), fourth-order polynomials are used in log-capital, log-investment, firm's export dummy, and import dummy to approximate g (.) .19 In addition, we also include a WTO dummy (i.e., one for a year after 2001 and zero for before) to characterize the function g (.) as follows: 4

4

g (kit , Iit , EFit ,WTOt )  (1  WTOt  EFit )  hq kith I itq

(5)

h 0 q 0

After finding the estimated coefficients ˆ m and ˆ l , we calculate the residual Rit which is defined as Rit  ln Yit  ˆ m ln Mit  ˆ l ln Lit The next step is to obtain an unbiased estimated coefficient of  k . We assume firm’s productivity follows a exogenous Markov process, vit  h(vit  1)   it .

To correct the selection

bias as mentioned above, Amiti and Konings (2007) suggested estimating the probability of a survival indicator on a high-order polynomial in log-capital and log-investment. One can then accurately estimate the following specification:

18

Olley and Pakes (1996) show that the investment demand function is monotonically increasing in the productivity shock vit , by making some mild assumptions about the firm's production technology. 19

Using higher order polynomials to approximate

g (.) 33

does not change the estimation results

Rit   k ln Kit  ?h gˆit  1   k

ˆ i , t  1)   it Ki , t  1 pr

(6)

ˆ i , t  1 denotes the fitted value for the probability of the firm 's exit in the next year., and where pr

 it   it   it denotes the composite error. Since the specific "true" functional form of the inverse function h is unknown, it is appropriate to use fourth-order polynomials in git  1 and

ln Ki , t  1 to approximate that. In addition, (6) also requires the estimated coefficients of the log-capital in the first and second term to be identical. Therefore, non-linear least squares seem to be the most desirable econometric technique (Pavcnik, 2002; Arnold, 2005). Finally, the Olley--Pakes type of TFP for each firm i in industry j is obtained once the estimated coefficient

ˆ k is obtained: TFPijtOP  ln Yit  ˆ m ln Mit  ˆ k ln Kit  ˆ l ln Lit

34

(7)